What is Artificial Intelligence?

Source: Deep Learning on Medium

What is Artificial Intelligence?

Defining some basic concepts

Image by Gerd Altmann from Pixabay

Given that Artificial Intelligence (AI) is quite the hype in modern times, one might be surprised to see a post defining AI. However, there are two reasons why this is important. One, because this article is the start of The Singularity, it is important for readers to know what is generally meant by AI in this publication. But maybe more importantly, there is still quite some confusion about what AI is in general. There are multiple terms in circulation, including Machine Learning and Deep Learning, and quite some people seem confused over how the terms are related. This post is an attempt to clear this up.

What does The Singularity mean by Artificial Intelligence?

As a publication, The Singularity uses quite a broad and clear definition of AI. First of all, AI is any intelligence demonstrated by machines instead of by biological entities like humans. Of course, this shifts the definition problem to “What is intelligence?” The Singularity uses the definition put forward by Shane Legg and Marcus Hutter in Universal Intelligence: A Definition of Machine Intelligence:

“Intelligence measures an agent’s ability to achieve goals in a wide range of environments.”

An agent is quite simply an entity that can monitor its environment and can act on it (change it) in a goal-oriented manner.

What about Machine Learning?

Most of the hype around AI in recent years is around Machine Learning (ML) and often even more specifically about a technique called Deep Learning. It seems as though ML is almost equated with AI, but really, ML is a subcategory of AI. As such, ML can always be viewed as AI, but the reverse is not necessarily true. So what exactly is ML?

Machine Learning happens when a machine, over a period of time, gets better at a task it wasn’t explicitly told how to do.

One could for example build an AI that learns to play the game of Chess by playing a (huge) number of Chess games against itself. By playing lots of Chess games, it will find patterns that lead to victory. Deep Learning is one technique to achieve this. How this works is a topic for an upcoming article in this publication, which will discuss ML in more detail and also explain Deep Learning.

Is That All?

No, it most certainly is not. AI is about much more than “just” ML. For example, even though it has been claimed that the World champion defeating Chess computer Deep Blue wasn’t a case of AI, according to the definition of intelligence given above, it most certainly was. Let’s repeat that definition:

“Intelligence measures an agent’s ability to achieve goals in a wide range of environments.”

Deep Blue had a goal: winning Chess games. Sure, Chess was the only environment Deep Blue operated in, and Deep Blue had but one goal, but it did have a huge ability to achieve that goal. Deep Blue, however, had the ability to play Chess from its beginning. It did not get better with playing. Therefore, it did not display Machine Learning. That doesn’t take anything away from its mastery of Chess though.

There is much more to say about Artificial Intelligence. We haven’t discussed various subtopics, or Artificial General Intelligence, Artificial Superintelligence, Friendly AI, the Intelligence Explosion or the Singularity. These last five terms in particular are of huge importance to this publication, and each deserves its own post.